Large Networks Research Articles

Control valve simulation method integrable into the global gradient algorithm

ABSTRACT This paper presents a refinement of a method that simulates flow- and pressure-regulating valves by replacing them with pipes and adjusting the resistances (diameters) of those pipes to meet the valve settings. The method, referred to as the replacing pipe method (RPM), is readily integrable into the global gradient algorithm, which is the heart of the commonly used EPANET 2 water distribution system modeling software. The RPM proved to be more robust, yet slower than the EPANET 2 valve handling heuristics. In this work, the RPM was modified to increase its computational speed. The modification involved the RPM check valve handling algorithm and condition for updating the diameters of the valve-replacing pipes. The modified RPM was tested as part of a hybrid simulator, which switches from the EPANET 2 valve handling heuristics to the RPM when the former fails. The test networks included a large network with thousands of nodes, pipes, and valves, and a case of an ill-posed problem. As the testing showed, the modified RPM offers a significant reduction in the number of iterations compared to its previous version.

Cluster-guided denoising graph auto-encoder for enhanced traffic data imputation and fault detection

In an era of increasing digitalization, the integration of diverse sensors has become commonplace. Within the transportation sector, state Departments of Transportation have extensively deployed traffic sensors to support various engineering endeavors. However, these sensors are prone to faults such as data loss, random noise, bias, and drift due to sensor aging, defects, or environmental influences. Therefore, detecting these faults is imperative to maintain data integrity. This paper presents a novel approach for detecting faults in traffic data based on a cluster-guided data reconstruction process. First, a traffic sensor clustering module is designed using a dual-encoding attention graph auto-encoder (DA-GAE) to identify clusters of traffic sensors. This module leverages a joint embedding of node and edge features in a low-dimensional vector space. Subsequently, utilizing the identified clusters, a cluster-guided denoising graph auto-encoder (CG-DGAE) is devised and trained for data reconstruction. The CG-DGAE employs a diffusion graph convolutional network (DGCN) and is trained with a cluster-wise sampling strategy. Extensive experiments were conducted using traffic data obtained from a real-world large sensor network, demonstrating superior performance of the CG-DGAE model in data reconstruction compared to various baseline methods. For fault detection, a score function is devised to discern potential faults by contrasting the sensor data sequence and the reconstructed data sequence. The fault detection results show that our proposed CG-DGAE model achieved an impressive overall accuracy of 99.09%, a precision of 99.13%, a recall of 99.53%, and a F1 score of 99.53%. The superior capability of CG-DGAE in data reconstruction and fault detection is attributable to the cluster-wise spatial–temporal context leveraged by the model.

Assessing the use of unstructured electronic health record data to identify exposure to firearm violence.

Research on firearm violence is largely limited to people who experienced acute bodily trauma and death which is readily gathered from Inpatient and Emergency Department settings and mortality data. Exposures to firearm violence, such as witnessing firearm violence or losing a loved one to firearm violence, are not routinely collected in health care. As a result, the true public health burden of firearm violence is underestimated. Clinical notes from electronic health records (EHRs) are a promising source of data that may expand our understanding of the impact of firearm violence on health. Pilot work was conducted on a sample of clinical notes to assess how firearm terms present in unstructured clinical notes as part of a larger initiative to develop a natural language processing (NLP) model to identify firearm exposure and injury in ambulatory care data. We used EHR data from 2012 to 2022 from a large multistate network of primary care and behavioral health clinics. A text string search of broad, gun-only, and shooting terms was applied to 9,598 patients with either/both an ICD-10 or an OCHIN-developed structured data field indicating exposure to firearm violence. A sample of clinical notes from 90 patients was reviewed to ascertain the meaning of terms. Among the 90 clinical patient notes, 13 (14%) notes reflect documentation of exposure to firearm violence or injury from firearms. Results from this study identified refinements that should be considered for NLP text classification. Unstructured clinical notes from primary and behavioral health clinics have potential to expand understanding of firearm violence.

Ethnoeconomy of Europe: Turkish Segment

The article examines the ethnic factor of the European economy. The migration of the Turks to Europe in the 1960s, formed a special segment of entrepreneurship in the region. A significant part of it was occupied by the representatives of the Turkish diaspora. Several factors including the social one contribute to the active development of the Turkish ethno-business in the European space. In addition, the formation and prosperity of the Turkish ethnic entrepreneurship in Europe is influenced by the high development of family and clan relations as well as the presence of various kinds of auxiliary mechanisms in the form of trade unions and business associations. In the context of small and medium-sized businesses, the key attention is paid to the development of the Turkish fast food industry related to the organization of food outlets, the hotel business, the provision of general services as well as tourism and logistics services. Large network business and transnational corporations of Turkey in most cases are engaged in such areas as: restaurant business; hotel business, including the construction of hotels; distribution of retail chains and Turkish brands. Islamic banking has also become a relatively new rapidly developing area of the Turkish ethnobusiness. The author concludes that the Turkish migration has significantly changed the nature of the European business. However, despite the wide representation of Turkish entrepreneurs in Europe, not all of them achieve the desired success and often face various kinds of discriminatory policies of the European authorities that restrict their activities.

Monte Carlo safeguarding of key links through multiple random walks in large network

Safeguarding important links is a useful way to promote network vulnerability, especially in sparse networks where random link failures can disconnect the network. Sustaining network connectivity is the main goal of safeguarding and can be handled by selectively safeguarding links belonging to certain cuts of a network. However, due to the inherent high complexity of cut enumeration, existing algorithms can only handle small‐scale networks with limited nodes. It is important to find practical algorithms that are feasible for large‐scale graphs, as a significant portion of real‐world graphs under investigation are not as small as dozens of nodes. To address the problem, we propose to use a high‐precision approximate approach to accelerate the first step of the two‐step safeguard process and thus accelerate the whole process. A Monte Carlo algorithm is proposed to exploit efficient random paths for small cuts enumeration, which can help locate important edges with high probability in large‐scale sparse networks. These edges will then be utilized to find the approximated minimum cost edge set whose safeguarding can sustain the expected network connectivity. The algorithm is validated using various sizes of graphs of random/Barbasi–Albert scale‐free/Clustered scale‐free/unit disk/Watts–Strogatz small‐world and real‐world graphs. The experimental results show that the algorithm can perfectly sustain the network connectivity, and the acceleration ratio of more than 105 can be achieved with a little additional overhead. Specifically, in large graphs of one million nodes, approximated safeguard solutions survive at least 99.9% random link failures, and the solution time can be further reduced to dozens of seconds when the algorithm steps are easily implemented in full parallel.

FELRec: efficient handling of item cold-start with dynamic representation learning in recommender systems

AbstractRecommender systems suffer from the cold-start problem whenever a new user joins the platform or a new item is added to the catalog. To address item cold-start, we propose to replace the embedding layer in sequential recommenders with a dynamic storage that has no learnable weights and can keep an arbitrary number of representations. In this paper, we present , a large embedding network that refines the existing representations of users and items in a recursive manner, as new information becomes available. In contrast to similar approaches, our model represents new users and items without side information and time-consuming finetuning, instead it runs a single forward pass over a sequence of existing representations. During item cold-start, our method outperforms similar method by 29.50–47.45%. Further, our proposed model generalizes well to previously unseen datasets in zero-shot settings. The source code is publicly available at https://github.com/kweimann/FELRec.

Prediction model optimization of gas turbine remaining useful life based on transfer learning and simultaneous distillation pruning algorithm

For the application of deep learning (DL) models in the field of remaining useful life (RUL) prediction and predictive maintenance (PdM) of complex equipment, the insufficient training data and large model are the two major problems. To address these issues, a model training method based on transfer learning and a simultaneous distillation pruning algorithm were proposed. By introducing prior knowledge, three transfer learning modes are devised to reduce the demand of training data. Additionally, the simultaneous distillation pruning algorithm was devised to make the model lightweight, and an iterative pruning method was adopted to trim the large neural network model. By analyzing the performance of different transfer learning modes, the effectiveness of the proposed method can be demonstrated. The number of model parameters and the performance before and after pruning were compared. The results demonstrated that, without significant alterations to the prediction performance, the proposed model exhibited the capability to markedly reduce the number of model parameters. Based on the proposed methods, the challenges of insufficient data and efficiency encountered by DL models could be effectively addressed.

Estimating electrical distribution network length and capital investment needs from real-world topologies and land cover data

Green technologies such as solar photovoltaic systems and electric vehicles play a fundamental role in the global decarbonization effort. To enable their diffusion, electricity distribution networks need to be upgraded, which is a complex and expensive endeavor. However, utilities face budgetary constraints and seek to reduce planning uncertainty. Here, we utilize 7,527 real-world grid topologies and land cover data to develop a model for estimating conductor and capital investment needs for electrifying a specific area. Our work yields three main contributions: First, we demonstrate the important role of land cover data in power line planning. We show that, for medium and large networks, distinct methodologies are needed due to the significant impact of land cover, particularly buildings and roads. Second, we introduce a parsimonious model of power line length and identify the number of consumption points as the primary determinant of network investment costs. Third, we present a cost assessment model tailored for regulators and investors, offering valuable insights for network planning, policymaking, due diligence, and research. Our work highlights the importance of combining land cover data and operations research algorithms in distribution network planning and provides policymakers with a tool to ensure cost-efficient network expansion.

Prediction of pre-eclampsia with machine learning approaches: Leveraging important information from routinely collected data

BackgroundGlobally, pre-eclampsia (PE) is a leading cause of maternal and perinatal morbidity and mortality. PE prediction using routinely collected data has the advantage of being widely applicable, particularly in low-resource settings. Early intervention for high-risk women might reduce PE incidence and related complications. We aimed to replicate our machine learning (ML) published work predicting another maternal condition (gestational diabetes) to (1) predict PE using routine health data, (2) identify the optimal ML model, and (3) compare it with logistic regression approach. MethodsData were from a large health service network with 48,250 singleton pregnancies between January 2016 and June 2021. Supervised ML models were employed. Maternal clinical and medical characteristics were the feature variables (predictors), and a 70/30 data split was used for training and testing the model. Predictive performance was assessed using area under the curve (AUC) and calibration plots. Shapley value analysis assessed the contribution of feature variables. ResultsThe random forest approach provided excellent discrimination with an AUC of 0.84 (95% CI: 0.82–0.86) and highest prediction accuracy (0.79); however, the calibration curve (slope of 1.21, 95% CI 1.13–1.30) was acceptable only for a threshold of 0.3 or less. The next best approach was extreme gradient boosting, which provided an AUC of 0.77 (95% CI: 0.76–0.79) and well-calibrated (slope of 0.93, 95% CI 0.85–1.01). Logistic regression provided good discrimination performance with an AUC of 0.75 (95% CI: 0.74–0.76) and perfect calibration. Nulliparous, pre-pregnancy body mass index, previous pregnancy with prior PE, maternal age, family history of hypertension, and pre-existing hypertension and diabetes were the top-ranked features in Shapley value analysis. ConclusionTwo ML models created the highest-performing prediction using routinely collected data to identify women at high risk of PE, with acceptable discrimination. However, to confirm this result and also examine model generalisability, external validation studies are needed in other settings, utilising standardised prognostic factors.

DOL3: Distilled OpenL3 audio embeddings for lightweight audio classification

Deep audio representations, also known as embeddings, recently became a popular alternative to conventional features like spectrograms for a wide range of audio classification tasks because of their domain-agnostic character and reduced training costs. Still, the usage is often limited to rather computationally intensive system due to the nature of their extraction from large networks. This paper aims to minimize the computational costs of embedding extraction by distilling the knowledge of the OpenL3 audio network to a smaller student network. Results show that the student network maintains comparable performance as the teacher network on various music and ambient noise classification tasks, while reducing the network size by over 90\% and the computational load by five times.

Dimensions: Calculating disruption indices at scale

Abstract Assessing the disruptive nature of a line of research is a new area of academic evaluation that moves beyond standard citation-based metrics by taking into account the broader citation context of publications or patents. The “CD index” and a number of related indicators have been proposed in order to characterize the disruptiveness of scientific publications or patents. This research area has generated a lot of attention in recent years, yet there is no general consensus on the significance and reliability of disruption indices. More experimentation and evaluation would be desirable, but it is hampered by the fact that the calculation of these indicators is time-consuming, especially if done at scale on large citation networks. We present a novel SQL-based method to calculate disruption indices for the Dimensions publications data on Google BigQuery. This reduces the computational time taken to produce such indices by an order of magnitude, as well as making available such functionalities within an online environment that requires no setup efforts. We explain the novel algorithm and describe how its results align with preexisting implementations of disruption indicators. This method will enable researchers to develop, validate, and improve disruption models more quickly and with more precision.

Nirsevimab Effectiveness Against Severe RSV Infection in the Primary Care Setting.

This study assesses the effectiveness of nirsevimab, a monoclonal antibody, in preventing medically attended respiratory syncytial virus-lower respiratory tract infections (RSV-LRTI) in a large primary care network in Spain, in both overall and catch-up infants under ten months of age. The 2023-24 immunization campaign with nirsevimab in Spain targeted all infants born after April 1st, 2023. Those born after October 1st received it at birth in hospitals, while others received it through a catch-up program. The MEDIPRIM network of primary care centers recruited all infants with LRTI for RSV PCR testing and employed a test-negative design (TND) approach to estimate the effectiveness of nirsevimab. The study included 160 infants, 141 (88%) of them received nirsevimab and 128 belonged to the catch-up group (88% received nirsevimab). Overall, RSV was detected in 44 infants (27.5%). Within the catch-up group, 37 (28.9%) were positive for RSV. The overall effectiveness was 75.8% (95% CI: 40.4-92.7), and 80.2% (95% CI: 44.3-95.4) in infants belonging to the catch-up group. This study underscores the effectiveness of nirsevimab in preventing medically attended LRTI in infants in outpatient settings and emphasizes the importance of a catch-up immunization program to reduce the disease burden in primary care.

Maintaining and updating accurate internal representations of continuous variables with a handful of neurons.

Many animals rely on persistent internal representations of continuous variables for working memory, navigation, and motor control. Existing theories typically assume that large networks of neurons are required to maintain such representations accurately; networks with few neurons are thought to generate discrete representations. However, analysis of two-photon calcium imaging data from tethered flies walking in darkness suggests that their small head-direction system can maintain a surprisingly continuous and accurate representation. We thus ask whether it is possible for a small network to generate a continuous, rather than discrete, representation of such a variable. We show analytically that even very small networks can be tuned to maintain continuous internal representations, but this comes at the cost of sensitivity to noise and variations in tuning. This work expands the computational repertoire of small networks, and raises the possibility that larger networks could represent more and higher-dimensional variables than previously thought.

In situ growth of large-scale and ultrathin carbon nitride films shield common metal substrates from corrosion and oxidation

A centimeter-size carbon nitride (CN) film, with nanoscale thickness, is in situ deposited on various metals through the thermal condensation of melamine, achieving broad and uniform coverage, without any pre- and post-treatments. As the temperature rises, the nitrogen content in the film increases, leading to the part of CN to transform into g-C3N4. Benefit from higher Gibbs adsorption free energies of g-C3N4 with Cl- and O2 in aqueous environment, a 30 nm CN film deposited at 475 ℃ with large g-C3N4 network, minimal segregation and low internal stress enables pure metals and alloys to achieve over 90 % protection efficiency.

B-114 Reference interval (RI) estimation for TSH in adults using an indirect method with unsupervised machine learning tool and R programming language

Abstract Background Thyroid-stimulating hormone (TSH) plays a fundamental role in regulating thyroid hormones T3 (triiodothyronine) and T4 (thyroxine). Understanding TSH test results is essential for effective management of thyroid disorders and depends, among other factors, on establishing specific reference intervals (RI) for the population being served. We aimed to estimate the reference intervals for TSH in the adult population served by a large diagnostic network in Brazil. Methods We employed data mining followed by estimation using an indirect method through the LabRI™ algorithm with parametric, non-parametric, and robust statistical treatments for outlier exclusion, a series of algorithms programmed in R Language. Included were TSH values from outpatients, aged 20 to 60 years, both sexes, from two analytical platforms. Used as a comparative RI recommended by the American Thyroid Association (ATA, 2014). Data were extracted from the Data Lake, from January to December 2022, from two laboratories in Brazil. Exclusion criteria: samples from hospital sources and referral laboratories; pregnant; some laboratory tests correlated with de TSH test; samples from patients using thyroid hormone and medications that may interfere with hormonal function and TSH laboratory analysis. Results After exclusion criteria, the sample totaled 26,909 adults of both sexes. After excluding 2.24% outliers, the estimated RI was 0.67-4.26 mIU/L, with the comparative RI chosen for this study being the one recommended by ATA of 0.40-4.00 mIU/L. The RI obtained in our study was similar to the Canadian RI harmonization study for TSH (0.60-4.48 mIU/L). Conclusions We estimated the RI, harmonized between Roche and Siemens platforms, to be 0.67-4.26 mIU/L. The RI obtained by the LabRI™ tool is equivalent to the Adeli study (2023) and is under discussion by the company's endocrinology committee for whether to use the estimated RI from our internal study or the one recommended by ATA (2014) in reports.

Multivariate and network lesion mapping reveals distinct architectures of domain-specific post-stroke cognitive impairments

BackgroundThe purpose of this study was to identify patterns of structural disconnection and multivariate lesion-behaviour relationships associated with post-stroke deficits across six commonly impacted cognitive domains: executive function, language, memory, numerical processing, praxis, and visuospatial attention. MethodsStroke survivors (n = 593) completed a brief domain-specific cognitive assessment (the Oxford Cognitive Screen (OCS)) during acute hospitalisation. Network-level and multivariate (sparce canonical correlation) lesion mapping analyses were conducted to identify focal neural correlates and distributed patterns of structural disconnection associated with impairment on each of the 16 OCS measures. ResultsNetwork-level and multivariate lesion mapping analyses identified significant correlates for 12/16 and 10/16 OCS measures, respectively which were largely consistent with correlates reported in past work. Language impairments were reliably localised to network- and voxel-level correlates centred in left fronto-temporal regions. Memory impairments were associated with disconnection in a large network of left hemisphere regions. Number processing deficits were associated with damage to voxels centred in the left insular/opercular cortex, as well as disconnection within the surrounding white matter tracts. Within the domain of attention, different subtypes of visuospatial neglect were linked to distinct but partially overlapping patterns of disconnection and voxel-level damage. Praxis impairment was not linked to any voxel-level regions but was significantly associated with disconnection within the left hemisphere dorsal attention network. ConclusionThese results highlight the utility of routine, domain-specific cognitive assessment and imaging data for theoretically-driven lesion mapping analyses, while providing novel insight into the complex anatomical correlates of common and debilitating post-stroke cognitive impairments.

B-120 Reference Interval (RI) Estimation for Pediatric Magnesium using an Indirect Method with Unsupervised Machine Learning Tool and Programming Language

Abstract Background Magnesium is an important intracellular component present in all enzymatic reactions involving Adenosine Triphosphate (ATP) utilization, with the majority stored in bones. In the blood, it exists as a free fraction and another bound to proteins. Regulation occurs through renal processes, and for this reason, hypermagnesemia is associated with renal insufficiency, compromising the central nervous system. On the other hand, hypomagnesemia cases are related to pancreatitis, intestinal malabsorption, childhood malnutrition, among other factors. Following complaints from pediatric doctors that our reference intervals (RI), as suggested by the manufacturer, would yield falsely elevated results in healthy patients, we estimated a new RI in the population served Methods We employed data mining followed by estimation using an indirect method through the LabRI™ algorithm with parametric, non-parametric, and robust statistical treatments for outlier exclusion, implemented in the R language. Latent abnormal values of outpatient pediatric patients were considered. We defined three pediatric partitions for both sexes, aged 0 to 6 years, 7 to 12 years, and above 12 years, using the comparative RI described by the Roche manufacturer. Data were extracted from our Data Lake between January and December 2022 in the Federal District of Brazil, from a laboratory part of a large network. Exclusion criteria included: Potassium > 5.5 and < 3.0 mmol/L; Phosphorus > 8.0 and < 1.5 mg/dL; Total Calcium > 12 mg/dL and < 7 mg/dL; Ionized Calcium > 1.5 mmol/L and < 0.8 mmol/L; Creatinine > 2.0 mg/dL; C-reactive protein (CRP) > 10 mg/dL; Ferritin > 700 ng/mL; A1C > 7.0%; LDL Cholesterol > 200 mg/dL; Magnesium repeated within < 12 months, and samples of hospital origin. Results The evaluated population after exclusion criteria totaled 6,865 children, distributed as follows: 26.0% aged 0 to 6 years, 30.3% aged 7 to 12 years, and 43.7% above 12 years. Outliers were excluded in 1.5%, 0.9%, and 1.0%, respectively. In all partitions, there was a significant discrepancy compared to the reference intervals recommended by Roche. In our estimation, there was an expansion of the upper limit of the reference interval in all partitions, supporting the clinical opinion. The new reference intervals became: 0-6 yrs: from 1.7-2.3 to 2.0-2.7; 7-12 yrs: from 1.7-2.1 to 2.0-2.6; and >12 yrs: from 1.7-2.2 to 1.9-2.6. The standard deviation varied between 0.19 and 0.20, and the median between 2.20 and 2.32 Conclusions The manufacturer states in the package insert that they did not assess the RI for the pediatric population and that each laboratory should verify the RI in the population served. We approve the adoption of the estimated RI for the pediatric population calculated by the LabRI™ tool in the partitions of 0-6 years, 7-12 years, and >12 years, stating in the report “Reference interval established by an indirect method in 6865 children tested in the laboratory, partitioned by age group.”

Analytical Bounds on the Local Lipschitz Constants of ReLU Networks.

In this article, we determine analytical upper bounds on the local Lipschitz constants of feedforward neural networks with rectified linear unit (ReLU) activation functions. We do so by deriving Lipschitz constants and bounds for ReLU, affine-ReLU, and max pooling functions, and combining the results to determine a network-wide bound. Our method uses several insights to obtain tight bounds, such as keeping track of the zero elements of each layer, and analyzing the composition of affine and ReLU functions. Furthermore, we employ a careful computational approach which allows us to apply our method to large networks such as AlexNet and VGG-16. We present several examples using different networks, which show how our local Lipschitz bounds are tighter than the global Lipschitz bounds. We also show how our method can be applied to provide adversarial bounds for classification networks. These results show that our method produces the largest known bounds on minimum adversarial perturbations for large networks such as AlexNet and VGG-16.

Scalability challenges of machine learning models for estimating walking and cycling volumes in large networks

This study explores the scalability of machine learning models for estimating walking and cycling volumes across the extensive New South Wales (NSW) Six Cities Region in Australia using mobile phone and crowdsourced data. Previous research has focused on localized applications, missing the complexities of larger networks. The research addresses this gap by identifying unique challenges such as the scarcity and representativeness of observed count data, gaps in the crowdsourced and mobile phone data, and inconsistencies in link-level volume estimates. We propose and demonstrate the application of strategies like enhancing geographical diversity of observed count data and employing an extensive cross-validation approach in model training and testing. By leveraging various auxiliary datasets, the study demonstrates the effectiveness of these strategies in improving model performance. These findings provide valuable insights for transportation modelers, policymakers, and urban planners, offering a robust framework for supporting sustainable transportation infrastructure and policies with advanced data-driven methodologies.

Dexamethasone Dose for COVID-19 in a Large U.S. Hospital Network From April 2020 to May 2023.

While potential harm from high doses of systemic dexamethasone for clinical management of COVID-19 is an important concern, little is known about real world dexamethasone dosing in patients hospitalized with COVID-19 in the United States. Descriptive study to assess dexamethasone daily dose in adults with COVID-19 in a large US hospital network, overall and by respiratory support requirements, extracted using semi- structured nursing notes. Of 332 430 hospitalizations with a COVID-19 diagnosis, 201 637 (60.7%) hospitalizations included dexamethasone administration. The mean age of recipients was 63 years, 53.0% were male, and 64.5% White. Median time from admission to dexamethasone administration was 0 day (interquartile range [IQR], 0-1 days) and median duration of use was 5 (IQR, 3-9) days. Almost 80% of hospitalizations received standard daily doses (≤ 6 mg daily), 12.7% moderately high daily doses (> 6- ≤ 10 mg daily), and 8.1% high (> 10- ≤ 20 mg daily) or very high daily dose (> 20 mg daily). Over 20% of COVID-19 hospitalizations requiring no oxygen or simple oxygen received high doses of systemic dexamethasone. Given the findings from the UK RECOVERY trial, and the general uncertainty around safety of higher dexamethasone doses in those requiring more intense respiratory support, standard daily dexamethasone doses of 6 mg or less for hospitalized COVID-19 requiring supplemental oxygen are recommended.